Merge commit 'e579d8b29cdb9b42c50a0fde277dfb047c1466ad'

[ffmpeg] / libavcodec / diracdsp.c

diff --git a/libavcodec/diracdsp.c b/libavcodec/diracdsp.c
index e82b58767eb4d025624602458844b7965a95f343..ab8d1497f717de7340ba6ce9dca4ba8522fa4343 100644 (file)
--- a/libavcodec/diracdsp.c
+++ b/libavcodec/diracdsp.c
@@ -20,7 +20,6 @@
 
 #include "avcodec.h"
 #include "diracdsp.h"
-#include "libavcodec/x86/diracdsp_mmx.h"
 
 #define FILTER(src, stride)                                     \
     ((21*((src)[ 0*stride] + (src)[1*stride])                   \
@@ -151,23 +150,28 @@ static void put_signed_rect_clamped_8bit_c(uint8_t *dst, int dst_stride, const u
     }
 }
 
-static void put_signed_rect_clamped_10bit_c(uint8_t *_dst, int dst_stride, const uint8_t *_src, int src_stride, int width, int height)
-{
-    int x, y;
-    uint16_t *dst = (uint16_t *)_dst;
-    int32_t *src = (int32_t *)_src;
-    for (y = 0; y < height; y++) {
-        for (x = 0; x < width; x+=4) {
-            dst[x  ] = av_clip(src[x  ] + 512, 0, (1 << 10) - 1);
-            dst[x+1] = av_clip(src[x+1] + 512, 0, (1 << 10) - 1);
-            dst[x+2] = av_clip(src[x+2] + 512, 0, (1 << 10) - 1);
-            dst[x+3] = av_clip(src[x+3] + 512, 0, (1 << 10) - 1);
-        }
-        dst += dst_stride >> 1;
-        src += src_stride >> 2;
-    }
+#define PUT_SIGNED_RECT_CLAMPED(PX)                                                                     \
+static void put_signed_rect_clamped_ ## PX ## bit_c(uint8_t *_dst, int dst_stride, const uint8_t *_src, \
+                                                  int src_stride, int width, int height)                \
+{                                                                                                       \
+    int x, y;                                                                                           \
+    uint16_t *dst = (uint16_t *)_dst;                                                                   \
+    int32_t *src = (int32_t *)_src;                                                                     \
+    for (y = 0; y < height; y++) {                                                                      \
+        for (x = 0; x < width; x+=4) {                                                                  \
+            dst[x  ] = av_clip_uintp2(src[x  ] + (1 << (PX - 1)), PX);                                  \
+            dst[x+1] = av_clip_uintp2(src[x+1] + (1 << (PX - 1)), PX);                                  \
+            dst[x+2] = av_clip_uintp2(src[x+2] + (1 << (PX - 1)), PX);                                  \
+            dst[x+3] = av_clip_uintp2(src[x+3] + (1 << (PX - 1)), PX);                                  \
+        }                                                                                               \
+        dst += dst_stride >> 1;                                                                         \
+        src += src_stride >> 2;                                                                         \
+    }                                                                                                   \
 }
 
+PUT_SIGNED_RECT_CLAMPED(10)
+PUT_SIGNED_RECT_CLAMPED(12)
+
 static void add_rect_clamped_c(uint8_t *dst, const uint16_t *src, int stride,
                                const int16_t *idwt, int idwt_stride,
                                int width, int height)
@@ -197,6 +201,7 @@ av_cold void ff_diracdsp_init(DiracDSPContext *c)
     c->add_rect_clamped = add_rect_clamped_c;
     c->put_signed_rect_clamped[0] = put_signed_rect_clamped_8bit_c;
     c->put_signed_rect_clamped[1] = put_signed_rect_clamped_10bit_c;
+    c->put_signed_rect_clamped[2] = put_signed_rect_clamped_12bit_c;
 
     c->add_dirac_obmc[0] = add_obmc8_c;
     c->add_dirac_obmc[1] = add_obmc16_c;
@@ -216,5 +221,6 @@ av_cold void ff_diracdsp_init(DiracDSPContext *c)
     PIXFUNC(avg, 16);
     PIXFUNC(avg, 32);
 
-    if (HAVE_MMX && HAVE_YASM) ff_diracdsp_init_mmx(c);
+    if (ARCH_X86)
+        ff_diracdsp_init_x86(c);
 }